Long Context Transfer from Language to Vision

TL;DR

This paper introduces a method to extend the context length of language models for vision tasks, enabling processing of extremely long videos without retraining, and demonstrates state-of-the-art results on a new long video benchmark.

Contribution

It proposes long context transfer by extrapolating language model context length to vision, and develops a synthetic benchmark to evaluate long video understanding.

Findings

Long Video Assistant processes over 200K visual tokens.

Achieves state-of-the-art on Video-MME with 7B-scale models.

Enables comprehension of 2000 frames without additional training.

Abstract

Video sequences offer valuable temporal information, but existing large multimodal models (LMMs) fall short in understanding extremely long videos. Many works address this by reducing the number of visual tokens using visual resamplers. Alternatively, in this paper, we approach this problem from the perspective of the language model. By simply extrapolating the context length of the language backbone, we enable LMMs to comprehend orders of magnitude more visual tokens without any video training. We call this phenomenon long context transfer and carefully ablate its properties. To effectively measure LMMs' ability to generalize to long contexts in the vision modality, we develop V-NIAH (Visual Needle-In-A-Haystack), a purely synthetic long vision benchmark inspired by the language model's NIAH test. Our proposed Long Video Assistant (LongVA) can process 2000 frames or over 200K visual tokens without additional complexities. With its extended context length, LongVA achieves state-of-the-art performance on Video-MME among 7B-scale models by densely sampling more input frames. Our work is open-sourced at https://github.com/EvolvingLMMs-Lab/LongVA.